Overlap Analysis

Compute pairwise overlap between two sets of detections.

Compute pairwise geometric overlap between two sets of detections.

For each pair (reference, candidate) drawn from reference_predictions x candidate_predictions, the block computes intersection_area / reference_polygon.area and emits one record per pair whose ratio reaches min_overlap.

When a detection carries a mask, the precise polygon is the longest contour of that mask (validated via shapely); otherwise the bounding box polygon is used. A vectorised bbox-IoU prefilter (supervision.box_iou_batch) eliminates non-touching pairs before any shapely intersection is computed.

The relation is intentionally not symmetric across the two inputs: the denominator is always the reference detection's area. Swap the two selectors if you want overlap reported relative to the other set.

The output is a flat list of dicts (one per accepted pair) attached to the same dimensionality as the inputs - the block does not increase dimensionality. See the detections_overlaps kind docs for the per-record schema.

Type identifier

Use the following identifier in step "type" field: roboflow_core/overlap_analysis@v1 to add the block as a step in your workflow.

Properties

Name	Type	Description	Refs
name	str	Enter a unique identifier for this step..	❌
min_overlap	float	Minimum (intersection / reference_area) ratio for a pair to be included in the output..	✅

The Refs column marks possibility to parametrise the property with dynamic values available in workflow runtime. See Bindings for more info.

Input and Output Bindings

The available connections depend on its binding kinds. Check what binding kinds Overlap Analysis in version v1 has.

Input and output bindings

• input

  • reference_predictions (Union[instance_segmentation_prediction, object_detection_prediction]): Detections whose area is the denominator of the overlap ratio. For each reference detection, overlap with every candidate is computed; pairs above min_overlap appear in the output..

  • candidate_predictions (Union[instance_segmentation_prediction, object_detection_prediction]): Detections checked against each reference..

  • min_overlap (float_zero_to_one): Minimum (intersection / reference_area) ratio for a pair to be included in the output..

• output

  • overlaps (detections_overlaps): List of per-pair detection overlap records.

Example JSON definition

{
	    "name": "<your_step_name_here>",
	    "type": "roboflow_core/overlap_analysis@v1",
	    "reference_predictions": "$steps.model_a.predictions",
	    "candidate_predictions": "$steps.model_b.predictions",
	    "min_overlap": 0.1
	}